Striping on Disks

Multimedia streams need continuous data supply. The aggregate data access requirement of many multimedia streams imposes very high demand on the access bandwidth of the storage servers. The disk striping or data striping methods spreads data over multiple disks to provide high aggregate disk throughput (Chua, Li, Ooi, & Tan, 1996; Hsieh, Lin, Liu, Du, & Ruwart, 1995). In addition to the popularity of multimedia objects that we have described in the last chapter, multimedia streams consume an object in a sequential manner. The striping methods make use of this access pattern to evenly spread the workload across disks. This can increase aggregate disk throughput so that high bandwidth streams can be delivered continuously. We first describe the simple striping method that places data stripes on a set of disks in the next section. After that, the staggered striping method that places data on a set of disks in a rotating manner is described. The pseudorandom placement method that stores data stripes on random disks is explained before we summarize this chapter.

Video Database Techniques and Video-on-Demand

Generally, a large-scale video server is composed of numerous disk striping groups. The striping policies employed by each disk striping group largely determine the performance of a video server. For storage and transmission efficiency, video data are usually compressed using variable-bit-rate (VBR) encoding algorithms, such as JPEG and MPEG. The amount of data consumed by a VBR video stream varies with time. This property, when coupled with striping, unfortunately, results in load imbalance across disks, degrading the overall server performance significantly. This chapter focuses on VBR video striping. It presents two state-of-the-art VBR striping schemes proposed in the literature: one is designed for homogeneous disks and the other is designed for heterogeneous disks. To gain insights into VBR striping, this chapter also develops performance models for the two striping policies. With these performance models, system designers can predict the maximum service capacity of a server, perform online admission control for clients, and optimize the performance of a server, without performing exhaustive tests on a real-system.

The Design of a Storage Server for Continuous Media

The increasing use of multimedia in computing is demanding more powerful computer hardware and system software. In particular, the stringent demands placed on multimedia file servers are unable to be satisfied by current magnetic disk technology or by current file systems techniques. This paper describes the design and implementation of a file server which has been specially optimized for continuous media. The critical performance issues which this paper addresses are the use of disk striping to provide high bandwidth, optimized disk layouts, hard real-time disk scheduling to ensure stream storage and retrieval guarantees are met, and a user interface to support high level multimedia abstractions.